A Crack Detection Method for Aero-engine Blade Based on Air-Flow Thermography

Aero-engine blade is one of the core components of aero-engine and its safe service is crucial to the normal operation of aero-engine. Thus, it is very important to localize and quantify cracks of aero-engine blades to prevent safety accidents. To guarantee the structural health of aero-engine blades, the air-flow thermography (AFT) method which has advantages of non-contact, high performance, and sensitivity, was proposed for detection of the blade cracks. Firstly, the theoretical thermal response model was established and the crack simulation models with different depths were constructed. Then a detection system was established to detect the artificial cracks with different depths and orientations. Results showed that the proposed method can effectively quantify the cracks with different depths from 0.2 mm to 1.0 mm and characterize the crack orientations. In addition, the natural crack on the blade was detected and the principal component analysis (PCA) method was used to enhance the crack contrast with the background.


Introduction
As the core power source of aircraft, the complex and precise structure of aero-engine is one of the standards of aircraft performance and national manufacturing technology [1]. The blade is the key component of aero-engine, and its structural integrity directly affects the performance and flight safety of the aircraft [2]. During the working process, the blades are always subjected to centrifugal loads caused by rotating motion, aerodynamic load and temperature load caused by engine operation, impact load, etc. In addition, the blades are usually affected by environmental erosion, resulting in foreign object damage (FOD) and erosion damages [3][4][5][6][7][8]. Therefore, it is susceptible to produce scratches and corrosion cracks on the surface of blades [9,10]. These failures are unexpected and catastrophic. In summary, the detection of aero-engine blade crack is an important guarantee for aircraft flight safety. Up to now, many detection techniques have been increasingly developed and applied to detect and maintain the aero-engine blades, such as penetrant testing, eddy current testing, ultrasonic testing, and borescope detection [11][12][13]. Fluorescent penetrant inspection (FPI) is widely used in NDT on aircraft parts, which is comparably more cost-efficient. Zheng et al. designed an advanced automatic inspection system for aircraft parts based on FPI [14]. But this method has a complicated operation process and can only detect the surface cracks. Furthermore, it will pollute the specimen and the environment. Phased array ultrasonic technology (PAUT) is a non-contaminant testing method, which is usually used in aerospace [15,16]. Xuegeng et al. applied the PAUT method on compressor impeller blade root crack and the defect location error is only 0.19% [17]. The PAUT is high accuracy and able to detect the subsurface cracks. However, it cannot detect the surface cracks due to the blind spot of ultrasonic detection method and the probe should be in contact with the blades. The eddy current testing (ECT) can detect both surface and the near-surface defects [18,19]. Sasi et al. used the dual-frequency eddy current technique to inspect the fatigue cracks of the aero-engines, and the results show that this method can detect fatigue cracks smaller than 2 mm [20]. Xia et al. designed and developed a small high sensitivity eddy current detection probe, which can be installed on the CNC multi-freedom scanning table and detect defects on the engine blades rapidly [21]. Lakshmi et al. used ECT to evaluate the stress-strain of aero-engine compressor stator vanes and their producer has been adopted for aero-engine during developmental phase [22]. However, with the above mentioned three methods, the probe needs to close contact with the blade in the detection process, resulting in complicated operation and time-consuming.
The bore-scope detection method can realize non-contact detection, and has been widely applied in defect detection and on-condition maintenance of aero-engine [23]. Yuan completed the inspection of the blade of CFM56-7B engine by using bore-scope, and determined the type of defect though the endoscope. However, the precision of bore-scope detection method always depends on the experience and patience of inspectors [24]. Guo approached a weighted morphology algorithm based on bore-scope detection, and He et al. proposed a network based on the Improved Cascade Mark R-CNN network to establish the damage related to the aeroengine blades and detection models [25,26]. However, it is still very difficult to detect the tiny cracks and scratches on the blade because the interior of the aero-engine is cramped and dim. Therefore, it's urgent to develop a non-contact method to detect the micro-cracks on the blades.
From the past researches, the air-flow heating method can realize long-distance thermal excitation. Meanwhile, the thermography method can get the temperature distribution from a distance and has been widely used in wind tunnels to test the stability of aircraft in the thermal environment [27][28][29]. The air-flow thermography (AFT) method which combines the advantages of the air-flow heating and the thermography can realize non-contact detection of the micro-cracks. Lu et al. proposed pulsed AFT for rail surface crack detection and evaluation [30]. The experimental results demonstrate that the AFT for rail surface cracks is effective at the low speed.
However, the mechanism of air-flow excitation is complicated, and the depth quantification and orientation characterization of the cracks on the blade are still the major challenges. In consideration of the above challenges, the theoretical model of the crack detection is established and the crack evaluation method based on the AFT is proposed in this paper. The rest sections of this paper are organized as follows: Section 2 establishes the temperature distribution model under air-flow heating which demonstrates that the AFT method can realize the cracks detection, and carries out the finite element simulation to determine the temperature distribution; Section 3 proposes the method to quantify the depths and orientations of artificial cracks, and the natural crack on the blade is detected; Section 4 draws the conclusion and proposes future works.

Principle of AFT
The heating processes of the specimen under the hot air-flow can be divided into two parts: thermal convection and thermal conduction. In this section, the temperature distribution model around the crack is established based on these two heating processes. Figure 1 shows the specimen with a crack in the hot air-flow field. The initial temperatures of the specimen and the hot air-flow are T sample and T flow , respectively. Due to the temperature difference between the specimen and the flow field (T flow >T sample ), the heat will flow from the high-temperature fluid to the low-temperature specimen. According to Newton's Law of Cooling, the thermal convection value can be calculated by: where, q denotes the thermal convection value, h is the convective heat-transfer coefficient, the temperature of the sample surface is T surface , and the temperature of the air-flow field is T flow .
As shown in Fig. 1, the side region of the defect could be regarded as a quarter-infinite body and the direction of heat convective can be divided into two directions during the thermal convection process. One is perpendicular to the plane S1 (yellow arrows) while another is perpendicular to the plane S2 (red arrows). Due to the stable hot air-flow during the heating process, the temperature and thermal convection coefficient remain constant. Therefore, the third boundary conditions are adopted as shown in Eqs. (2 and 3) [31].
where, h 1 denotes the thermal convection coefficient between the hot air-flow and the plane S 1 , h 2 denotes the thermal convection coefficient between the hot air-flow and the plane S 2 . According to Fourier's law of thermal conduction, the thermal differential equation of transient temperature field T (x, y, z, T ) in solid heat conduction without internal heat source is: where, λ x , λ y , λ z , are the thermal conductivity of the material along the x, y and z directions, respectively; T denotes the temperature; ρ is the density; c represents the thermal capacity. The Eq. (4) is solved according to the product theorem to obtain the temperature distribution around the crack [32]. The temperature distribution can be expressed as Eq. (5): where α λ/(ρc) is the thermal diffusion coefficient of the material.
In practice, the thermal convection coefficient is affected by the roughness and the state of fluid, which leads to the thermal convection coefficient at the cracks, is different from that at the sound area [33]. Usually, the roughness of crack area is bigger than that in the sound area. Due to airflow disturbances in the crack area, the heat convection direction at plane S 1 is parallel to flow direction of hot air-flow, while the heat convection direction at plane S 2 is perpendicular to flow direction of hot air-flow. Thus the thermal convection coefficient at plane S 1 (h 1 ) is larger than that at plane S 2 (h 2 ). It can be represented as: In Fig. 1, part A and part B are captured to represent the crack area and sound area. It can be seen that part A is heated by the two planes and the part B is heated only by the plane S 2 . Furthermore, the energy of S 1 through the thermal convection process is larger than that at S 2 . Thus, the temperature distribution can be expressed by: From the above analysis, when the blade is heated by the air-flow, the temperature at the crack area will be significantly higher than that at sound area. Therefore, the crack can be detected because of the abnormal thermal response by using AFT.

Verify the Theory with Simulation
To verify the theory and investigate the heating process and results of engine blade crack excited by hot air-flow, a 3D simulation model was established as shown in Fig. 2a based on COMSOL. The dimensions of the specimens were 150 × 100 × 3 mm 3 and the material was titanium as pictured in  The hot air-flow has a temperature of 500°C and a velocity of 10 m/s. The specimen was heated continuously for 5 s. The velocity and direction of the air-flow inside the crack on the cross-section plane are shown in Fig. 3. The velocity of the air-flow inside the crack is smaller than that in the sound area and the direction of the air-flow at the crack area is perpendicular to the plane. Figure 4a is the temperature distribution of the specimens with different crack depths after heating for 5 s and Fig. 4b is the temperature of l 1 . It shows that the temperature values along the length direction of cracks are much higher than that of sound areas which is coincident with the theoretical analysis. In addition, the temperature variations at points (A, B, C, D, E) are captured and shown in Fig. 4c, and the results show that the temperature along the crack length is positively correlated with the crack depth. Figure 4d shows that the temperature after 5 s heating time (captured from the black dotted box in Fig. 4c) beside the crack area represents a monotonic relation with the crack depths. Therefore, the simulation results demonstrate that the AFT can be used for crack detection and depth quantification.

AFT Setup
The AFT setup is shown in Fig. 5a. The system mainly consists of four parts: an excitation source with a hot air gun, a pulse controller, an infrared (IR) camera and a PC. In this work, the pulse controller is built on STM32F407 and the heating duration is 5 s; the maximum power of the hot air gun is 2.0 kW; the temperature and the flow velocity are 500°C and 500 L/min respectively. The IR camera is FLIR A655sc with the frame rate of 200 Hz and the resolution of 640 × 120 pixels. Specifically, the size of the plate was 150 × 100 × 3 mm 3 and the artificial cracks have the constant width (0.2 mm) and length (10 mm) but varied depths from 0.2 to 1.0 mm with an interval of 0.2 mm. As shown in Fig. 6b, the materials of the specimen are titanium alloy. Before conducting the experiments, the top faces of all specimens were coated with matte black paint to increase the emissivity.

Depth Quantification
The specimen was continuously heated 5 s and the temperature changes of the specimen were captured by the IR camera. The temperature distribution at the region of interest (ROI) with different crack depths after 5 s heating time is shown in Fig. 6a. The result demonstrates that the temperature at the crack area is higher than that at the sound area. It is consistent with theoretical analysis and simulation results. Hence, the AFT method can be used to detect the cracks on the blades. The average temperature of the 3 points with different crack depths are extracted in Fig. 6b and T 1 is the heating period. This work proposes two features to characterize crack depths. The first feature is peak temperature after the heating process at l 1 . Figure 6c gives the relationship between the temperature at l 1 and the different crack depths. The solid curve presents its linear fit. From this figure, it can be see the temperature at the crack region after 5 s heating time represents a monotonic relation with crack depths. The R 2 and RMSE of the fitted line are 99.64% and 0.06. Another feature to achieve depths characterization is the slopes of the heating curve. The fitting results of the heating curves are shown in Fig. 6d. The R 2 and RMSE of the fitted line are listed in Table 1. Figure 6e shows the relationship between the slopes of the heating curve and the crack depths. The R 2 and RMSE of the fitted line are 98.87% and 0.0001. Hence, based on the linear relationship with crack depths, the peak temperature after the heating process and the slope of the heating curve   can be selected as effective features to quantify the crack depths.

Orientation Characterization
To characterize the crack orientations, the cracks with varied orientations from 15°to 90°with a 15°interval are detected. The crack depth is 0.6 mm. The temperature distributions after 5 s and the results after the principal component analysis (PCA) are shown in Fig. 7. It can be seen that the temperature at the crack area is higher than that at the sound area in the raw images. And the temperature difference is enhanced at the PCA results. This work proposes a coordinate calculation method to characterize orientations. Firstly, the grey values Fig. 7 The temperature distribution at the ROI after 5 s heating time and PCA results with different orientations.

Fig. 8
The grey value at l 1 and l 2 at l 1 and l 2 are extracted and shown in Fig. 8. It is easy to find that the grey values at the crack area are obviously lower than that at the sound area. Therefore, the coordinate of the points with minimum values are chosen. As shown in Fig. 7, the orientation of the connecting line between points A and B can be approximately taken as the orientation of the crack. Therefore, based on the coordinate values of the two points, Fig. 9 The detection results of the crack orientation the orientation of the crack can be calculated by: where, θ denotes the orientation of the crack; x and y is the difference of the coordinate between point A and B. Based on the above method, the calculation results are fitted in Fig. 9. The R 2 and RMSE of the fitted line are 98.32%

Detection of Natural Crack
To verify the detection effect of this method on cracks, the detection experiments of natural crack are carried out, as shown in Fig. 10. The engine blade is nearly 130 mm × 40 mm × 170 mm (as shown in Fig. 10a) and natural crack is approximately 0.1 mm × 0.1 mm × 70 mm (as shown in Fig. 10b). The experimental setup and parameters are the same as those in Sect. 3.1. The hot air-flow is generated by the hot air gun with the maximum power of 2.0 kW. The temperature and the flow velocity are 500°C and 500 L/min respectively. The temperature distribution at the crack area after 5 s heating is captured by the IR camera as pictured in Fig. 10c. The temperature at the crack area is apparently higher than that at the sound area, and the crack at part A is more obvious than that in Fig. 10b. To enhance the crack area, the PCA method was used and the result is shown in Fig. 10d. By comparing the two parts of Fig. 10c and d, it is evident that the PCA-based analysis can effectively extract and enhance the crack area, which proves that the AFT method has a good effect on the detection of micro cracks on the blades.

Conclusion
The AFT technique was introduced for crack detection and depth quantification in aluminum alloy material. Results showed that peak temperature after the heating process and the slope of the heating curve can be used to quantify the crack depths within 1 mm. And the approximate connecting line is used to characterize the crack orientation. Additionally, AFT can be used to detect the natural micro crack on the engine blade and the PCA method can be used to enhance the crack detection result. Future work will focus on applying AFT to characterize the natural micro crack on the engine blade. Furthermore, feature selection will also be investigated to characterize different geometric parameters.